标记错分样本的AdaBoost算法

AdaBoost algorithm with wrongly classified samples marked

提出一种新的标记迭代过程中错分样本的AdaBoost算法(MWBoost),该算法通过在提升过程中,把上一个分类器错分的样本全部参入到下一个分类器的训练中,并在分类正确的样本中进行重采样,从而使得后一轮提升中分类器能够更快速地关注那些难以分类的样本。该算法在UCI的多个数据集上进行了测试,并且与传统的AdaBoost算法进行了比较,实验结果表明,新的算法具有更好的分类精度。

A new training method of AdaBoost（MWBoost） which marks samples wrongly classified in the former iteration is proposed.Using the method of re-sampling in the samples those correctly classified and putting all the wrongly classified in the next training, the classification can pay more attention on those samples hardly to classify.The algorithm is tested on the UCI benchmark data sets and compared with the original AdaBoost algorithm;the result shows that MWBoost performs better than AdaBoost with more accuracy.